Speaker
Description
The work proposed in this presentation is part of a global project dealing
with the characterisation of heterogeneous media by both electromagnetic and
mechanical full waveform inversion. Indeed Full Waveform Inversion (FWI) of
seismic reflection (SR) or Ground Penetrating Radar (GPR) data is an efficient
approach to reconstruct subsurface physical parameters with high resolution.
We focus here on the mechanical part and more specifically on quantitative
imaging of nearsurface density in the context of water content characterization,
as analogous to relative dielectric permittivity involved in the electromagnetical
approach. Processing field data is challenging because the nature of the source
and the sensors used impact the signal-to-noise ratio as well as the frequency
range embedded in the recorded data. From then it becomes interesting to
process the data in the frequency domain, by working on a few representative
frequencies of the recorded temporal signal. Field data are simulated by noisy
synthetic data. Different frequency strategies are used and their results are
compared with each other. The inverse problem involved consists in assessing
the density in the probed medium, from the data on the displacement field mea-
sured at the detectors. Such a problem is known to be nonlinear and ill-posed.
It is solved iteratively by a regularized Gauss-Newton algorithm (RGN), which
relies on the Fréchet derivative obtained through the generalized reciprocity
principle which is equivalent to the well-known adjoint method. The numerical
results obtained show an optimal strategy, for which the convergence rate and
the computation time are reasonable, the spatial resolution is improved and the
density is well reconstructed.
